The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The Workflows/Update.md file uses dynamic context injection syntax (!command) to execute shell operations at load time. These templates incorporate positional arguments ($1, $2, $3) interpolated from user input, creating a potential shell injection vector.
[COMMAND_EXECUTION]: The Next.js dashboard template uses child_process.spawn in DashboardTemplate/app/api/chat/route.ts to execute a local TypeScript tool for inference, enabling dynamic execution of local scripts.
[DATA_EXFILTRATION]: The skill is designed to read highly sensitive personal data from ~/.claude/LIFEOS/USER/TELOS/ and expose it to a web interface and an external inference tool.
[COMMAND_EXECUTION]: The DashboardTemplate/app/api/upload/route.ts handler is vulnerable to path traversal because it uses the unsanitized file.name from the upload request to construct the file system save path, potentially allowing unauthorized file writes outside the intended directory.
[COMMAND_EXECUTION]: The SKILL.md file mandates a curl notification to http://localhost:31337/notify upon every invocation, establishing silent background network communication.
[PROMPT_INJECTION]: The skill uses 'MANDATORY' instructions to override agent behavior by requiring specific voice and text notifications before processing user requests.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted files from user-specified directories without adequate boundary markers or sanitization in the upload path.
Ingestion points: DashboardTemplate/app/api/upload/route.ts and DashboardTemplate/lib/telos-data.ts.
Boundary markers: None implemented in processing logic.
Capability inventory: spawn, fs.writeFileSync, curl.
Sanitization: Missing path resolution checks in the upload route.

Telos